Choosing a Cycle Count Cadence for a Mid-Market Apparel Brand

It is a Tuesday in October at a $15M contemporary brand. The DTC team just oversold a bestseller by 42 units because the 3PL feed lagged four hours behind Shopify. The wholesale ops lead is on Slack asking why the ATS in the B2B portal shows 180 units of a style the warehouse insists it shipped last week. Someone is in a spreadsheet reconciling three inventory sources. The founder wants to know why this keeps happening after they paid for a WMS. Nobody has run a cycle count in six weeks because nobody agreed whose job it was.

What is a cycle count cadence for an apparel brand?

A cycle count cadence apparel operations teams can defend is a written schedule that says which SKUs get counted, in which locations, at what frequency, by whom, and what happens when the count disagrees with the system. It is not an annual wall-to-wall. It is not a vibe. It is a matrix that treats inventory accuracy as a maintained state rather than a periodic audit.

For a mid-market apparel brand running wholesale plus DTC plus a 3PL, the cadence has three axes: SKU velocity, location risk, and event triggers. Velocity decides how often a SKU is counted in steady state. Location risk decides which bins and zones need more frequent eyes regardless of what sits in them. Event triggers decide when the schedule gets interrupted by reality, a drop launch, a return wave, a peak promo, a 3PL variance report that does not net to zero.

This lives inside Breakpoint 3 of the 6 Breakpoints framework, where inventory truth gets weaker. BP3 is not caused by lazy counting. It is caused by cadence design that does not match how apparel actually moves.

Why does apparel inventory drift faster than other categories?

Apparel has structural drift sources that hard-goods categories do not. Size and color proliferation means one style can be 40 to 80 SKUs, and a mispick between size M and size L rarely triggers a return until the customer opens the box. Drops and preorders create allocation pools that sit in the system as committed inventory but physically live in the same bins as available inventory. Returns come back in waves, sit on a QC bench for days, and post to available stock on a lag. Wholesale ships in cases and DTC ships in eaches from the same physical inventory, and the unit-of-measure conversions are a rich source of ghost stock.

Layer a 3PL on top and you get a second system of record that reconciles nightly at best. When I am sitting across from a buyer comparing vendors, the question that separates a serious operator from a hopeful one is not “how often should we count?” It is “where does your system drift first, and how do you know?” Teams that cannot answer that question do not need a better WMS. They need a cadence.

The back-of-envelope pattern at $15M is consistent: 6 to 9 hours a week burned reconciling Shopify, the 3PL, and wholesale, a 2 to 3 percent oversell rate at peak, and one FTE effectively doing data plumbing instead of operations. A cycle count cadence does not eliminate that work. It converts it from firefighting into a maintained process.

How do you build the cadence by SKU velocity?

Start with an ABC classification refreshed every quarter, not annually. In apparel, a style that was C-velocity in July can be A-velocity in October because a stylist wore it. Any classification older than a season is fiction.

A-velocity SKUs are the top 20 percent of units shipped in the trailing 90 days. These get counted weekly in the pick face. If a SKU generates a third of your DTC volume and half of your wholesale allocation, a two-week drift window is unacceptable. Weekly counts on A-velocity typically take a picker 30 to 45 minutes for a mid-market catalog, and they catch mispicks before they compound into oversells.

B-velocity SKUs are the next 30 percent. Monthly counts are the floor. If your B tier drifts, your open-to-buy math for the next season drifts with it, because B is where you are testing what wants to become A.

C-velocity SKUs, slow movers, carryover, and archive, get counted quarterly. The temptation is to count them less because they “do not move.” That is exactly why they drift, because nobody looks at them and the bin becomes a graveyard for putaway errors.

Reserve and overflow locations get counted quarterly regardless of velocity, and every time a pick pulls from reserve instead of pick face, the reserve bin gets a count within seven days. That last rule catches more variance than any weekly schedule, because reserve is where the 3PL forgets what it has.

How do location risk and channel exposure change the cadence?

Velocity is not the only axis. A B-velocity SKU that is allocated against a wholesale PO with a hard ship window in nine days is temporarily an A. A C-velocity carryover style that is featured in next week’s email drop is temporarily an A. Cadence has to react to commitment, not just historical throughput.

This is where channel exposure matters. A brand running wholesale plus DTC on the same inventory pool has to decide how ATS is calculated per channel. Wholesale should not run through Shopify’s native flow, and it should not share an uncommitted ATS with DTC. If your B2B portal and your DTC storefront both see 100 units and both can sell them, you will oversell. The cadence that supports channel-aware ATS is a weekly count on every SKU that has an open wholesale allocation, because the cost of a chargeback on a short-ship is higher than the cost of the count.

Location risk adds a third layer. Any bin that has been touched by a returns processor in the last 30 days gets a count on a compressed schedule. Any bin that shares a shelf with a lookalike SKU, same style, adjacent size or colorway, gets counted whenever its neighbor is counted. Any bin that the 3PL flagged in a nightly variance report gets a count within 48 hours, no exceptions.

When should event-driven counts override the schedule?

There are five events that should interrupt the regular cadence. Drop launches trigger a full count of every SKU in the drop 72 hours before go-live, because a drop that oversells is a customer service event and a refund event and a brand event stacked on top of each other. This is the Magnolia Pearl pattern for teams running same-day fulfillment on drops with international duty complexity, where a mis-count on drop day cascades into duty reclassification and returns math that takes weeks to unwind.

Return waves trigger counts on the affected SKUs within 72 hours of the return posting to inventory. Returns should post to inventory in days, not weeks, and the count confirms the posting is real. If your returns take three weeks to hit available stock, your cadence is compensating for a workflow problem, and no count frequency will fix it.

Peak promo events, Black Friday, sample sale, wholesale market week, trigger daily spot counts on the top 50 SKUs by projected volume for the duration of the event. Not full cycle counts. Spot counts on pick face, five minutes per SKU.

3PL variance reports that do not net to zero trigger a count within 48 hours on every SKU in the report. If the 3PL is telling you they lost 12 units and found 12 units of a different SKU, that is a pick error pattern, not a math error, and it will keep happening until you count.

System migrations and physical moves trigger a full wall-to-wall on the affected zones. This is the only place an old-school wall-to-wall still earns its keep.

What does the cadence look like in practice at $15M?

For a brand doing $15M with roughly 1,200 to 2,000 active SKUs, one 3PL, wholesale plus DTC, the weekly rhythm looks like this. Monday morning: a picker spends 90 minutes on A-velocity pick face counts, roughly 60 to 80 SKUs. Tuesday: 3PL variance report from the weekend gets reviewed, any non-zero lines trigger counts by Wednesday. Wednesday: B-velocity rolling counts, one-fourth of the B tier each week so the full B tier is covered monthly. Thursday: any SKU on an open wholesale allocation with a ship window inside 14 days gets counted. Friday: reserve location spot checks on anything picked from reserve that week.

Quarterly, the C tier gets a full pass over two weeks, and the ABC classification gets refreshed against trailing 90-day velocity. Annually, there is no wall-to-wall, because the maintained cadence made it unnecessary and a wall-to-wall costs a weekend of overtime and produces a snapshot that is stale by Tuesday.

What I see from prospects who have already shortlisted three vendors is that they usually have some version of this written down, and none of it is enforced, because the system they are using cannot tie a count to a variance to a root cause. The cadence lives in a Notion page. The counts live in a spreadsheet. The variance analysis lives in someone’s head. That is why BP3 gets worse instead of better even as the team adds process.

How do you know the cadence is working?

Three metrics, tracked weekly. Inventory record accuracy at the SKU-location level, measured as the percentage of counted lines where system quantity matches physical quantity within a tolerance of zero for A tier and one unit for B and C. Target is 98 percent for A, 95 percent for B, 92 percent for C. Below those, the cadence is not tight enough or the workflow feeding the system has a leak.

Oversell rate, measured as units oversold divided by units shipped, tracked per channel. A $15M brand at peak sits in the 2 to 3 percent range without a cadence. A defensible cadence pulls that under 1 percent and under 0.5 percent outside of peak. If your oversell rate is not moving after 90 days of the cadence, the problem is not counting. It is channel-aware ATS, and no count will fix a data model that lets two channels sell the same unit.

Hours per week spent on inventory reconciliation, tracked honestly. The 6 to 9 hour range at $15M should compress to 2 to 3 hours within a quarter of running a real cadence, because the count is doing the reconciliation instead of a spreadsheet doing it after the fact.

Where does the cadence break, and what does that mean?

Cadences break in predictable places. They break when the counter and the person who resolves variance are the same person, because that person will resolve toward the system rather than toward reality. Separate the roles. They break when variance thresholds are set by dollar value instead of unit count, because a 40-unit variance on a $12 tee is a bigger operational problem than a 2-unit variance on a $400 jacket, even though the jacket variance is bigger in dollars. They break when the 3PL runs its own cadence in parallel and neither side reconciles to the other, which is not a cadence problem, it is a contract problem.

They also break when the underlying system cannot represent a count in a way that closes the loop. If your count sheet is a CSV export and your adjustment is a manual entry in a separate screen, the cadence will decay within two quarters because the friction is too high. A cadence is only as durable as the workflow that captures it.

What this means for an apparel operations team

A cycle count cadence is not a warehouse hygiene practice. It is the mechanism by which BP3 gets held in place while the rest of the operation scales. If inventory truth is weak, every downstream number, open-to-buy, gross margin, chargeback exposure, allocation confidence, is a guess dressed up as a report.

The teams that get this right stop treating counts as a chore and start treating them as the sensor network for the operation. Variance is signal. A weekly count that finds nothing is not wasted time. It is confirmation that the picking, putaway, returns, and allocation workflows upstream are behaving. The moment a count starts finding variance, the cadence has done its job by telling you which workflow to look at.

The design question is not “how often do we count?” It is “what does the count tell us, and can we act on it before the next drop?” A brand that can answer that question in one connected system has closed the BP3 loop. A brand that cannot is going to spend another quarter reconciling three sources and calling it operations.